Human-Content-to-Machine-Data_Final - Flipbook - Page 32
academic researchers, who would not have the resources to instead rely on costly licensing
deals.
In early workshops about this work, we brainstormed a wide range of tactics we could use to
stimulate further demand. On the positive side, we could develop a CC-designed badge, a
certiûcation system, or other methods for AI developers who adhere to CC signals to use to
indicate their participation in this prosocial system. Alternatively, or perhaps in parallel, we
could leverage more aggressive tactics like publicly identifying AI developers who do not
adhere to CC signals.
Working in Tandem and in Partnership
CC signals is one element of the path towards a more equitable future.
As we9ve acknowledged before, preference signals are by themselves not sufûcient to help
sustain the commons,104 and other interventions will be required to grow it going forward.
There remains much work to do to increase transparency around the data used to train
models, as well as to reinforce the technical infrastructure that underpins large collections of
content on the web. Wikimedia Enterprise, for example, demonstrates how developing
commercial-grade APIs for Wikipedia and other sources of knowledge can help ensure that
high-quality open data can be sustained for everyone.105
We9re also conscious that CC signals won9t address broader issues concerning the
distribution and use of data. These issues, such as risks to data protection and safety, must
be tackled by other means, including by choosing not to share something at all if the risk of
misuse or abuse is particularly high.106
Safeguarding What Matters
Let9s not forget what we are protecting. A thriving creative commons belongs to all of us,
including humans using machines to generate insights and discoveries. But a creative
commons will not thrive on extraction or neglect. It requires care, reciprocity, and intention.
Call it stewardship, a circular economy, or regeneration, the principle is the same: the
commons must be replenished by the collective it serves.
104
Timid Robot. (2023, 31 August). Exploring Preference Signals for AI Training - Creative Commons.
Creative Commons.
https://creativecommons.org/2023/08/31/exploring-preference-signals-for-ai-training/
105
Wikimedia Enterprise. (n.d.). Learn more about Wikimedia Enterprise. Wikimedia Enterprise.
https://enterprise.wikimedia.com/about/
106
Downing, K. (2023, July 13). AI Licensing Can9t Balance
